Efficient-UCBV: An Almost Optimal Algorithm Using Variance Estimates

Authors: Subhojyoti Mukherjee, K. P. Naveen, Nandan Sudarsanam, Balaraman Ravindran

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through an extensive numerical study we show that EUCBV significantly outperforms the popular UCB variants (like MOSS, OCUCB, etc.) as well as Thompson sampling and Bayes-UCB algorithms.
Researcher Affiliation Academia 1Department of Computer Science & Engineering, Indian Institute of Technology Madras 2Department of Electrical Engineering, Indian Institute of Technology Tirupati 3Department of Management Studies, Indian Institute of Technology Madras 4 Robert Bosch Centre for Data Science and AI (RBC-DSAI), Indian Institute of Technology Madras
Pseudocode Yes Algorithm 1 EUCBV
Open Source Code No The paper does not provide a link or explicit statement about the open-sourcing of the EUCBV algorithm's code. It mentions code for other algorithms: 'The implementation for KLUCB, Bayes-UCB and DMED were taken from Cappe, Garivier, and Kaufmann (2012)'.
Open Datasets No The paper describes experiments using synthetically generated data (e.g., '20 Bernoulli distributed arms', '100 arms involving Gaussian reward distributions') rather than named public datasets with explicit access information.
Dataset Splits No The paper does not explicitly describe train/validation/test dataset splits. For multi-armed bandit problems, the algorithm learns sequentially rather than on pre-split datasets in the manner of supervised learning.
Hardware Specification No No specific hardware (e.g., GPU/CPU models, memory) used for running the experiments is mentioned in the paper.
Software Dependencies No The paper mentions that implementations for some baseline algorithms were taken from a citation ('Cappe, Garivier, and Kaufmann (2012)') but does not list specific software dependencies with version numbers for their own algorithm or experimental setup.
Experiment Setup Yes The parameters of EUCBV algorithm for all the experiments are set as follows: ψ = T K2 and ρ = 0.5 (as in Corollary 1). Experiment-1 (Bernoulli with uniform gaps): This experiment is conducted to observe the performance of EUCBV over a short horizon. The horizon T is set to 60000. The testbed comprises of 20 Bernoulli distributed arms with expected rewards of the arms as r1:19 = 0.07 and r 20 = 0.1...